cmvr-es/test/e2e/run_cmvr_es_quic_smoke.py

#!/usr/bin/env python3
"""Run a no-hardware cmvr_es process against the local MsQuic test gateway."""

from __future__ import annotations

import argparse
import json
import signal
import subprocess
import tempfile
import time
from pathlib import Path


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    parser.add_argument("--gateway", type=Path, required=True)
    parser.add_argument("--cmvr-es", type=Path, required=True)
    parser.add_argument("--cert", type=Path, required=True)
    parser.add_argument("--key", type=Path, required=True)
    parser.add_argument("--timeout-seconds", type=float, default=15.0)
    return parser.parse_args()


def wait_for_json(
    path: Path, process: subprocess.Popen[str], deadline: float
) -> dict[str, object]:
    last_error: Exception | None = None
    while time.monotonic() < deadline:
        if process.poll() is not None:
            raise RuntimeError(
                f"gateway exited before publishing {path.name}: "
                f"returncode={process.returncode}"
            )
        if path.is_file():
            try:
                return json.loads(path.read_text(encoding="utf-8"))
            except (OSError, json.JSONDecodeError) as error:
                last_error = error
        time.sleep(0.02)
    raise TimeoutError(
        f"timed out waiting for {path}"
        + (f": {last_error}" if last_error else "")
    )


def stop_process(process: subprocess.Popen[str] | None) -> int | None:
    if process is None:
        return None
    if process.poll() is not None:
        return process.returncode
    process.send_signal(signal.SIGTERM)
    try:
        process.wait(timeout=3.0)
    except subprocess.TimeoutExpired:
        process.kill()
        process.wait(timeout=3.0)
    return process.returncode


def write_runtime_config(config_dir: Path, port: int) -> Path:
    config_dir.mkdir(parents=True, exist_ok=True)
    (config_dir / "cmvr_es.pb.txt").write_text(
        """
cmvr_es {
  logger_config_file: "logger.pb.txt"
  device_manager_config_file: "device_manager.pb.txt"
  task_manager_config_file: "task_manager.pb.txt"
}
""".lstrip(),
        encoding="utf-8",
    )
    (config_dir / "logger.pb.txt").write_text(
        """
logger {
  minimum_level: LOG_LEVEL_INFO
  routes { level: LOG_LEVEL_INFO terminal: true }
  routes { level: LOG_LEVEL_WARNING terminal: true }
  routes { level: LOG_LEVEL_ERROR terminal: true }
  routes { level: LOG_LEVEL_FATAL terminal: true }
  max_file_size_mb: 1
  flush_interval_seconds: 1
  format {
    show_time: false
    show_level: true
    show_thread_id: false
    show_source_location: true
  }
}
""".lstrip(),
        encoding="utf-8",
    )
    (config_dir / "device_manager.pb.txt").write_text(
        """
device_manager {
  name: "cmvr-quic-process-smoke"
  version: "test"
  description: "no-hardware QUIC process smoke test"
  devices {
    id: "disabled-smoke-camera"
    type: DEVICE_TYPE_CAMERA
    enable: false
  }
  devices {
    id: "disabled-smoke-microphone"
    type: DEVICE_TYPE_MICROPHONE
    enable: false
  }
  devices {
    id: "enabled-missing-camera"
    type: DEVICE_TYPE_CAMERA
    enable: true
  }
}
""".lstrip(),
        encoding="utf-8",
    )
    (config_dir / "task_manager.pb.txt").write_text(
        """
task_manager {
  tasks {
    id: "quic_edge"
    type: TASK_TYPE_QUIC_EDGE
    run_mode: TASK_RUN_MODE_BLOCKING_SERVICE
    config_file: "quic_edge.pb.txt"
    enable: true
  }
}
""".lstrip(),
        encoding="utf-8",
    )
    (config_dir / "quic_edge.pb.txt").write_text(
        f"""
quic_edge {{
  id: "quic_edge"
  enable: true
  server_host: "127.0.0.1"
  server_port: {port}
  alpn: "cmvr-quic-edge/1"
  node_id: "cmvr-process-smoke"
  software_version: "test"
  grpc_endpoint_host: "127.0.0.1"
  grpc_endpoint_port: 50052
  grpc_endpoint_tls: false
  include_loopback_interfaces: true
  heartbeat_interval_ms: 250
  control_response_timeout_ms: 1000
  tls {{ allow_insecure: true }}
  reconnect {{
    initial_delay_ms: 50
    maximum_delay_ms: 250
    multiplier: 2.0
    jitter_percent: 0
    connect_timeout_ms: 2000
  }}
  maximum_datagram_bytes: 1200
  maximum_control_frame_bytes: 1048576
  maximum_frame_bytes: 16384
  datagram_send_queue_depth: 64
  media_poll_interval_ms: 2
}}
""".lstrip(),
        encoding="utf-8",
    )
    return config_dir / "cmvr_es.pb.txt"


def require_count(summary: dict[str, object], key: str, minimum: int) -> None:
    value = summary.get(key)
    if not isinstance(value, int) or value < minimum:
        raise RuntimeError(
            f"gateway summary {key}={value!r}, expected at least {minimum}"
        )


def run() -> int:
    args = parse_args()
    for path in (args.gateway, args.cmvr_es, args.cert, args.key):
        if not path.is_file():
            raise FileNotFoundError(path)

    deadline = time.monotonic() + args.timeout_seconds
    gateway_process: subprocess.Popen[str] | None = None
    edge_process: subprocess.Popen[str] | None = None

    with tempfile.TemporaryDirectory(prefix="cmvr-es-quic-smoke-") as temp:
        temp_dir = Path(temp)
        ready_file = temp_dir / "ready.json"
        summary_file = temp_dir / "summary.json"
        gateway_log = temp_dir / "gateway.log"
        edge_log = temp_dir / "cmvr_es.log"

        try:
            with gateway_log.open("w", encoding="utf-8") as gateway_output:
                gateway_process = subprocess.Popen(
                    [
                        str(args.gateway),
                        "--bind",
                        "127.0.0.1",
                        "--port",
                        "0",
                        "--cert",
                        str(args.cert),
                        "--key",
                        str(args.key),
                        "--heartbeat-interval-ms",
                        # A zero response keeps the edge-side value below,
                        # proving heartbeat_interval_ms is configurable.
                        "0",
                        "--ready-file",
                        str(ready_file),
                        "--summary-file",
                        str(summary_file),
                        "--exit-after-heartbeats",
                        # The production client keeps only one heartbeat
                        # outstanding. Receiving heartbeat 3 therefore proves
                        # that ACKs 1 and 2 were processed by cmvr_es.
                        "3",
                    ],
                    stdout=gateway_output,
                    stderr=subprocess.STDOUT,
                    text=True,
                )
                ready = wait_for_json(ready_file, gateway_process, deadline)
                port = ready.get("port")
                if not isinstance(port, int) or not 0 < port <= 65535:
                    raise RuntimeError(f"invalid gateway ready payload: {ready}")

                root_config = write_runtime_config(temp_dir / "config", port)
                with edge_log.open("w", encoding="utf-8") as edge_output:
                    edge_process = subprocess.Popen(
                        [str(args.cmvr_es), str(root_config)],
                        stdout=edge_output,
                        stderr=subprocess.STDOUT,
                        text=True,
                    )
                    remaining = max(0.1, deadline - time.monotonic())
                    gateway_returncode = gateway_process.wait(timeout=remaining)
                    if gateway_returncode != 0:
                        raise RuntimeError(
                            f"gateway exited with {gateway_returncode}"
                        )

                    edge_returncode = stop_process(edge_process)
                    edge_process = None
                    if edge_returncode != 0:
                        raise RuntimeError(
                            f"cmvr_es exited with {edge_returncode}"
                        )

            summary = json.loads(summary_file.read_text(encoding="utf-8"))
            if summary.get("runtime_failed") is not False:
                raise RuntimeError(f"gateway runtime failure: {summary}")
            require_count(summary, "registrations_accepted", 1)
            require_count(summary, "heartbeats_received", 3)
            require_count(summary, "heartbeat_acks_sent", 3)
            require_count(summary, "registration_interface_count", 1)
            require_count(summary, "heartbeat_interface_count", 1)
            if summary.get("heartbeat_has_device_manager") is not True:
                raise RuntimeError(
                    f"DeviceManager snapshot was not received: {summary}"
                )
            if (
                summary.get("heartbeat_device_manager_name")
                != "cmvr-quic-process-smoke"
                or summary.get("heartbeat_device_manager_version") != "test"
                or summary.get("heartbeat_device_count") != 1
                or summary.get("heartbeat_enabled_device_count") != 1
                or summary.get("heartbeat_disabled_device_count") != 0
                or summary.get("heartbeat_error_device_count") != 1
                or summary.get("heartbeat_unknown_health_device_count") != 1
            ):
                raise RuntimeError(
                    f"unexpected DeviceManager heartbeat snapshot: {summary}"
                )
            devices = summary.get("heartbeat_devices")
            if not isinstance(devices, list) or len(devices) != 1:
                raise RuntimeError(
                    f"heartbeat device rows are missing: {summary}"
                )
            row = devices[0]
            if (
                not isinstance(row, dict)
                or row.get("device_id") != "enabled-missing-camera"
                or row.get("kind") != 5
                or row.get("type_name") != "Camera"
                or row.get("enabled") is not True
                or row.get("manager_state") != 7
                or row.get("health") != 0
                or row.get("has_error") is not True
                or row.get("error_message") != "device creation failed"
                or not isinstance(
                    row.get("status_updated_at_unix_ms"), int
                )
                or row["status_updated_at_unix_ms"] <= 0
            ):
                raise RuntimeError(
                    f"unexpected heartbeat device row: {row!r}"
                )
            if summary.get("protocol_violations") != 0:
                raise RuntimeError(f"protocol violation: {summary}")
            if summary.get("last_node_id") != "cmvr-process-smoke":
                raise RuntimeError(f"unexpected registered node: {summary}")
            if (
                summary.get("last_grpc_endpoint_host") != "127.0.0.1"
                or summary.get("last_grpc_endpoint_port") != 50052
                or summary.get("last_observed_source_ip") != "127.0.0.1"
            ):
                raise RuntimeError(
                    f"IP/gRPC endpoint report was not observed: {summary}"
                )

            edge_output_text = edge_log.read_text(encoding="utf-8")
            if "[QuicEdgeTask] Started" not in edge_output_text:
                raise RuntimeError(
                    "cmvr_es did not start the QUIC task:\n" + edge_output_text
                )
            if "CMVR_HAS_MSQUIC is not compiled" in edge_output_text:
                raise RuntimeError(
                    "cmvr_es used the unavailable QUIC stub:\n"
                    + edge_output_text
                )

            print(
                "cmvr_es_quic_process_smoke_test: PASS "
                + json.dumps(summary, sort_keys=True)
            )
            return 0
        except Exception:
            if gateway_log.is_file():
                print("gateway log:\n" + gateway_log.read_text(encoding="utf-8"))
            if edge_log.is_file():
                print("cmvr_es log:\n" + edge_log.read_text(encoding="utf-8"))
            raise
        finally:
            stop_process(edge_process)
            stop_process(gateway_process)


if __name__ == "__main__":
    raise SystemExit(run())